required_packages <- c("dplyr", "ggplot2", "knitr", "maps", "tidyr")
missing_packages <- required_packages[!vapply(required_packages, requireNamespace, logical(1), quietly = TRUE)]

if (length(missing_packages) > 0) {
  stop(
    "Install the missing packages before knitting: ",
    paste(missing_packages, collapse = ", ")
  )
}

library(dplyr)
library(ggplot2)
library(knitr)
library(maps)
library(tidyr)

knitr::opts_chunk$set(
  echo = TRUE,
  message = FALSE,
  warning = FALSE,
  fig.width = 11,
  fig.height = 7,
  fig.align = "center"
)

dir.create("visuals", showWarnings = FALSE, recursive = TRUE)
dir.create("data/processed", showWarnings = FALSE, recursive = TRUE)

theme_set(theme_minimal(base_size = 12))

1 1. Project overview

1.1 1.1 Background

Childhood and adolescent obesity is a major public-health challenge because it affects physical and mental health, quality of life and later noncommunicable-disease risk. The sixth round of the WHO European Childhood Obesity Surveillance Initiative (COSI) found substantial differences between countries and renewed concern about the lack of progress after the COVID-19 period. The 2021/2022 Health Behaviour in School-aged Children (HBSC) report additionally shows that weight status, diet and physical activity are socially patterned.

The 18th European Public Health Conference 2025 abstract supplement sharpened the policy context for this project. Its childhood-obesity contributions highlighted four connected challenges: interventions limited to one behaviour or setting; unequal food and activity environments; weak protection from unhealthy food marketing; and the need for standardized, measurement-based surveillance. The report therefore treats obesity as a population and systems problem rather than an individual failure.

The present dataset can quantify differences by country, sex and time. It cannot directly measure family affluence, food marketing, school meals or policy implementation. Those determinants are used to interpret the findings and define future data extensions, not as causes tested in this analysis.

1.2 1.2 Primary research question

How large and persistent are inequalities in modelled obesity prevalence among children and adolescents aged 5–19 years across the WHO European Region from 1990 to 2022, considering country, sex, change over time and uncertainty?

1.3 1.3 Secondary research questions

Which countries have the highest latest prevalence, and how wide are their confidence intervals?
Which countries experienced the largest absolute and annualized changes between their first and latest estimates?
Did the spread between lower- and higher-prevalence countries narrow or widen over time?
How large is the male–female prevalence gap in each country, and did the average absolute gap change over time?
How should the patterns be interpreted alongside current European concerns about socioeconomic disadvantage, unhealthy food environments, physical inactivity and food marketing?

The first four questions are answered quantitatively. The fifth is a contextual interpretation and does not claim causal effects.

1.4 1.4 Objectives

Validate and clean the WHO-derived dataset.
Describe its geographic and temporal coverage.
Compare the latest country estimates with confidence intervals.
Quantify absolute, relative and annualized country-level change.
Measure between-country inequality using the interquartile range and P90–P10 gap.
Quantify the direction and magnitude of sex differences over time.
Separate observed inequalities from unmeasured explanations and policy determinants.
Export an analysis-ready file for Tableau.

1.5 1.5 Unit of analysis and terminology

Each row represents one country–year–sex–age-group estimate. Results are ecological country-level estimates and must not be interpreted as individual-level risk.

In this report, inequality means a measurable difference between countries or sex categories. Inequity implies that a difference is avoidable and unfair. Establishing inequity requires information on social position and structural determinants that is not present in the core dataset.

1.6 1.6 Analytical framework

analytical_framework <- data.frame(
  Domain = c(
    "Current burden", "Long-term change", "Geographic inequality",
    "Sex inequality", "Estimate uncertainty", "Systems context"
  ),
  Measure = c(
    "Latest prevalence and rank", "Absolute, relative and annualized change",
    "IQR and P90–P10 gap", "Male minus female gap and absolute gap",
    "95% confidence interval width", "Narrative triangulation only"
  ),
  Interpretation = c(
    "Where modelled prevalence is highest", "Direction and speed of change",
    "Whether country distributions converge or diverge",
    "Direction and magnitude of sex difference",
    "Precision of each modelled estimate",
    "Food, activity, school, community and policy environment"
  )
)

kable(analytical_framework, caption = "Conference-informed analytical framework")

Conference-informed analytical framework
Domain	Measure	Interpretation
Current burden	Latest prevalence and rank	Where modelled prevalence is highest
Long-term change	Absolute, relative and annualized change	Direction and speed of change
Geographic inequality	IQR and P90–P10 gap	Whether country distributions converge or diverge
Sex inequality	Male minus female gap and absolute gap	Direction and magnitude of sex difference
Estimate uncertainty	95% confidence interval width	Precision of each modelled estimate
Systems context	Narrative triangulation only	Food, activity, school, community and policy environment

2 2. Data source and variables

The raw file is stored at data/raw/childhood_obesity.csv. It contains modelled estimates from the WHO indicator Prevalence of obesity among children and adolescents aged 5–19 years. The source URL is retained in every row for provenance.

data_dictionary <- data.frame(
  Variable = c(
    "Country", "Country_code", "Year", "Sex", "Age_group",
    "Obesity_prevalence", "Lower_CI", "Upper_CI", "Source"
  ),
  Meaning = c(
    "Country name", "ISO alpha-3 country code", "Estimate year",
    "Female, Male or Total", "WHO age-group code",
    "Modelled obesity prevalence (%)", "Lower confidence limit (%)",
    "Upper confidence limit (%)", "WHO indicator URL"
  )
)

kable(data_dictionary, caption = "Data dictionary")

Data dictionary
Variable	Meaning
Country	Country name
Country_code	ISO alpha-3 country code
Year	Estimate year
Sex	Female, Male or Total
Age_group	WHO age-group code
Obesity_prevalence	Modelled obesity prevalence (%)
Lower_CI	Lower confidence limit (%)
Upper_CI	Upper confidence limit (%)
Source	WHO indicator URL

3 3. Import and validation

3.1 3.1 Read the CSV

Using a relative path makes the project reproducible on another computer after cloning the repository.

data_path <- "/Users/test/Desktop/Research with Codex/childhood_obesity.csv"

if (!file.exists(data_path)) {
  stop("Data file not found: ", data_path)
}

obesity_raw <- read.csv(
  data_path,
  na.strings = c("", "NA"),
  check.names = FALSE,
  stringsAsFactors = FALSE
)

required_columns <- c(
  "Country", "Country_code", "Year", "Sex", "Age_group",
  "Obesity_prevalence", "Lower_CI", "Upper_CI", "Source"
)

missing_columns <- setdiff(required_columns, names(obesity_raw))
if (length(missing_columns) > 0) {
  stop("Missing required columns: ", paste(missing_columns, collapse = ", "))
}

3.2 3.2 Clean and derive analysis variables

This step converts numeric fields, creates readable labels and removes rows that cannot contribute to the analysis. The original raw CSV remains unchanged.

obesity <- obesity_raw |>
  transmute(
    Country = trimws(Country),
    Country_code = toupper(trimws(Country_code)),
    Year = as.integer(Year),
    Sex = trimws(Sex),
    Age_group = trimws(Age_group),
    Age_group_label = recode(
      Age_group,
      "Y5T9" = "5–9 years",
      "Y10T19" = "10–19 years",
      "Y5T19" = "5–19 years",
      .default = Age_group
    ),
    Obesity_prevalence = as.numeric(Obesity_prevalence),
    Lower_CI = as.numeric(Lower_CI),
    Upper_CI = as.numeric(Upper_CI),
    Source = trimws(Source)
  ) |>
  filter(
    Country != "",
    !is.na(Year),
    !is.na(Obesity_prevalence)
  ) |>
  arrange(Country, Year, Sex, Age_group)

kable(
  head(obesity, 10),
  digits = 2,
  caption = "First ten cleaned observations"
)

First ten cleaned observations
Country	Country_code	Year	Sex	Age_group	Age_group_label	Obesity_prevalence	Lower_CI	Upper_CI	Source
Albania	ALB	1990	Female	Y10T19	10–19 years	1.28	0.35	3.06	https://data.who.int/indicators/i/C6262EC/EF93DDB
Albania	ALB	1990	Female	Y5T19	5–19 years	1.69	0.49	3.89	https://data.who.int/indicators/i/C6262EC/EF93DDB
Albania	ALB	1990	Female	Y5T9	5–9 years	2.45	0.74	5.54	https://data.who.int/indicators/i/C6262EC/EF93DDB
Albania	ALB	1990	Male	Y10T19	10–19 years	2.24	0.66	5.24	https://data.who.int/indicators/i/C6262EC/EF93DDB
Albania	ALB	1990	Male	Y5T19	5–19 years	3.32	1.09	7.49	https://data.who.int/indicators/i/C6262EC/EF93DDB
Albania	ALB	1990	Male	Y5T9	5–9 years	5.29	1.78	11.46	https://data.who.int/indicators/i/C6262EC/EF93DDB
Albania	ALB	1990	Total	Y10T19	10–19 years	1.77	0.75	3.49	https://data.who.int/indicators/i/C6262EC/EF93DDB
Albania	ALB	1990	Total	Y5T19	5–19 years	2.52	1.10	4.79	https://data.who.int/indicators/i/C6262EC/EF93DDB
Albania	ALB	1990	Total	Y5T9	5–9 years	3.91	1.76	7.26	https://data.who.int/indicators/i/C6262EC/EF93DDB
Albania	ALB	1991	Female	Y10T19	10–19 years	1.32	0.39	3.05	https://data.who.int/indicators/i/C6262EC/EF93DDB

4 4. Data-quality assessment

4.1 4.1 Coverage summary

coverage_summary <- data.frame(
  Measure = c(
    "Rows", "Countries", "First year", "Latest year",
    "Sex categories", "Age groups", "Duplicate keys"
  ),
  Value = c(
    nrow(obesity),
    n_distinct(obesity$Country),
    min(obesity$Year),
    max(obesity$Year),
    paste(sort(unique(obesity$Sex)), collapse = ", "),
    paste(sort(unique(obesity$Age_group_label)), collapse = ", "),
    obesity |>
      count(Country, Year, Sex, Age_group) |>
      filter(n > 1) |>
      nrow()
  )
)

kable(coverage_summary, caption = "Dataset coverage and uniqueness checks")

Dataset coverage and uniqueness checks
Measure	Value
Rows	15444
Countries	52
First year	1990
Latest year	2022
Sex categories	Female, Male, Total
Age groups	10–19 years, 5–19 years, 5–9 years
Duplicate keys	0

4.2 4.2 Missingness and logical checks

Confidence limits should contain the point estimate. Any violations are counted so that they are visible rather than silently ignored.

quality_checks <- data.frame(
  Check = c(
    "Missing prevalence", "Missing lower confidence limit",
    "Missing upper confidence limit", "Prevalence outside 0–100%",
    "Lower limit above estimate", "Upper limit below estimate"
  ),
  Count = c(
    sum(is.na(obesity$Obesity_prevalence)),
    sum(is.na(obesity$Lower_CI)),
    sum(is.na(obesity$Upper_CI)),
    sum(obesity$Obesity_prevalence < 0 | obesity$Obesity_prevalence > 100, na.rm = TRUE),
    sum(obesity$Lower_CI > obesity$Obesity_prevalence, na.rm = TRUE),
    sum(obesity$Upper_CI < obesity$Obesity_prevalence, na.rm = TRUE)
  )
)

kable(quality_checks, caption = "Missingness and logical validation")

Missingness and logical validation
Check	Count
Missing prevalence	0
Missing lower confidence limit	0
Missing upper confidence limit	0
Prevalence outside 0–100%	0
Lower limit above estimate	0
Upper limit below estimate	0

5 5. Descriptive analysis

5.1 5.1 Define the latest comparable population

For a fair country comparison, the analysis uses the latest year shared by the data and restricts results to total-sex estimates for ages 5–19 years.

latest_year <- max(obesity$Year, na.rm = TRUE)

latest_total <- obesity |>
  filter(
    Year == latest_year,
    Sex == "Total",
    Age_group == "Y5T19"
  ) |>
  mutate(
    Burden_quartile = factor(
      ntile(Obesity_prevalence, 4),
      levels = 1:4,
      labels = c("Low", "Moderate", "High", "Very high")
    )
  )

latest_summary <- latest_total |>
  summarise(
    Year = latest_year,
    Countries = n_distinct(Country),
    Mean = mean(Obesity_prevalence, na.rm = TRUE),
    Median = median(Obesity_prevalence, na.rm = TRUE),
    Minimum = min(Obesity_prevalence, na.rm = TRUE),
    Maximum = max(Obesity_prevalence, na.rm = TRUE)
  )

kable(latest_summary, digits = 1, caption = "Latest comparable prevalence summary")

Latest comparable prevalence summary
Year	Countries	Mean	Median	Minimum	Maximum
2022	52	8.4	8.6	1.6	14.8

5.2 5.2 Highest-prevalence countries

highest_prevalence <- latest_total |>
  slice_max(Obesity_prevalence, n = 15, with_ties = FALSE) |>
  arrange(desc(Obesity_prevalence))

kable(
  highest_prevalence |>
    select(Country, Year, Obesity_prevalence, Lower_CI, Upper_CI, Burden_quartile),
  digits = 1,
  caption = "Fifteen highest latest estimates, total population aged 5–19"
)

Fifteen highest latest estimates, total population aged 5–19
Country	Year	Obesity_prevalence	Lower_CI	Upper_CI	Burden_quartile
Hungary	2022	14.8	12.4	17.4	Very high
Cyprus	2022	14.1	10.5	18.0	Very high
Finland	2022	12.6	10.1	15.5	Very high
Malta	2022	12.1	8.8	15.8	Very high
Bulgaria	2022	12.1	9.5	14.7	Very high
Bosnia and Herzegovina	2022	12.0	6.7	18.6	Very high
Türkiye	2022	11.7	7.9	16.0	Very high
Greece	2022	11.6	9.2	14.1	Very high
North Macedonia	2022	11.3	8.2	15.1	Very high
United Kingdom of Great Britain and Northern Ireland	2022	11.3	9.4	13.4	Very high
Andorra	2022	11.2	3.9	21.3	Very high
Serbia	2022	10.9	8.2	13.9	Very high
Slovenia	2022	10.8	9.8	12.0	Very high
Estonia	2022	10.7	8.9	12.6	High
Georgia	2022	10.6	7.3	14.7	High

ranking_plot <- highest_prevalence |>
  mutate(Country = reorder(Country, Obesity_prevalence)) |>
  ggplot(aes(x = Obesity_prevalence, y = Country)) +
  geom_col(fill = "#C94C4C", width = 0.72) +
  geom_errorbar(
    aes(xmin = Lower_CI, xmax = Upper_CI),
    orientation = "y",
    width = 0.22,
    colour = "#333333"
  ) +
  labs(
    title = paste("Highest modelled obesity prevalence in", latest_year),
    subtitle = "Total population aged 5–19; lines show confidence intervals",
    x = "Obesity prevalence (%)",
    y = NULL,
    caption = "Source: WHO"
  )

ranking_plot

ggsave("visuals/latest_obesity_prevalence.png", ranking_plot, width = 11, height = 7, dpi = 200)

6 6. Inequality analysis by sex

The sex gap is defined as male prevalence minus female prevalence. Positive values indicate a higher estimate among males; negative values indicate a higher estimate among females.

sex_gap_all <- obesity |>
  filter(
    Age_group == "Y5T19",
    Sex %in% c("Female", "Male")
  ) |>
  select(Country, Country_code, Year, Sex, Obesity_prevalence) |>
  pivot_wider(names_from = Sex, values_from = Obesity_prevalence) |>
  mutate(
    Sex_gap = Male - Female,
    Absolute_sex_gap = abs(Sex_gap)
  ) |>
  filter(!is.na(Sex_gap)) |>
  arrange(Country, Year)

sex_gap <- sex_gap_all |>
  filter(Year == latest_year) |>
  arrange(desc(abs(Sex_gap)))

kable(
  head(sex_gap, 15),
  digits = 1,
  caption = "Largest absolute sex gaps in the latest year"
)

Largest absolute sex gaps in the latest year
Country	Country_code	Year	Female	Male	Sex_gap	Absolute_sex_gap
Serbia	SRB	2022	7.4	14.3	6.9	6.9
Montenegro	MNE	2022	5.2	12.0	6.8	6.8
Georgia	GEO	2022	7.6	13.4	5.8	5.8
Cyprus	CYP	2022	11.2	16.8	5.7	5.7
Finland	FIN	2022	10.0	15.2	5.3	5.3
Estonia	EST	2022	8.1	13.2	5.1	5.1
Greece	GRC	2022	9.0	14.0	5.0	5.0
Bosnia and Herzegovina	BIH	2022	9.5	14.5	4.9	4.9
Hungary	HUN	2022	12.3	17.2	4.9	4.9
Malta	MLT	2022	9.6	14.4	4.8	4.8
Romania	ROU	2022	6.5	11.1	4.6	4.6
Slovenia	SVN	2022	8.5	13.1	4.6	4.6
Sweden	SWE	2022	7.8	12.3	4.5	4.5
Italy	ITA	2022	7.4	11.7	4.3	4.3
North Macedonia	MKD	2022	9.1	13.4	4.3	4.3

sex_gap_plot <- sex_gap |>
  slice_max(abs(Sex_gap), n = 15, with_ties = FALSE) |>
  mutate(Country = reorder(Country, Sex_gap)) |>
  ggplot(aes(x = Sex_gap, y = Country, fill = Sex_gap > 0)) +
  geom_col(show.legend = FALSE) +
  geom_vline(xintercept = 0, colour = "#333333") +
  scale_fill_manual(values = c("TRUE" = "#3B78A8", "FALSE" = "#D9795B")) +
  labs(
    title = paste("Largest sex differences in", latest_year),
    subtitle = "Male prevalence minus female prevalence, ages 5–19",
    x = "Percentage-point difference",
    y = NULL
  )

sex_gap_plot

ggsave("visuals/latest_sex_gap.png", sex_gap_plot, width = 11, height = 7, dpi = 200)

6.1 6.1 Change in the regional sex gap

The mean absolute gap describes the typical size of the country-level difference regardless of which sex has the higher estimate. The mean signed gap retains direction.

sex_gap_trend <- sex_gap_all |>
  group_by(Year) |>
  summarise(
    Mean_signed_gap = mean(Sex_gap, na.rm = TRUE),
    Mean_absolute_gap = mean(Absolute_sex_gap, na.rm = TRUE),
    Countries = n_distinct(Country),
    .groups = "drop"
  )

sex_gap_trend_plot <- sex_gap_trend |>
  select(Year, `Signed gap` = Mean_signed_gap, `Absolute gap` = Mean_absolute_gap) |>
  pivot_longer(-Year, names_to = "Measure", values_to = "Gap") |>
  ggplot(aes(Year, Gap, colour = Measure)) +
  geom_hline(yintercept = 0, colour = "grey65") +
  geom_line(linewidth = 1) +
  scale_colour_manual(values = c("Signed gap" = "#3B78A8", "Absolute gap" = "#C94C4C")) +
  labs(
    title = "Country-level sex inequality over time",
    subtitle = "Unweighted mean across countries, ages 5–19",
    x = "Year", y = "Percentage-point gap", colour = NULL
  )

sex_gap_trend_plot

ggsave("visuals/sex_gap_trend.png", sex_gap_trend_plot, width = 11, height = 7, dpi = 200)

7 7. Time trends and between-country inequality

7.1 7.1 Regional descriptive trend

The regional trend is an unweighted mean of available country estimates. It describes the average country trajectory and is not a population-weighted European prevalence estimate.

regional_trend <- obesity |>
  filter(Age_group == "Y5T19", Sex %in% c("Female", "Male", "Total")) |>
  group_by(Year, Sex) |>
  summarise(
    Mean_prevalence = mean(Obesity_prevalence, na.rm = TRUE),
    Countries = n_distinct(Country),
    .groups = "drop"
  )

trend_plot <- ggplot(
  regional_trend,
  aes(x = Year, y = Mean_prevalence, colour = Sex)
) +
  geom_line(linewidth = 1) +
  scale_colour_manual(values = c("Female" = "#D979A7", "Male" = "#3B78A8", "Total" = "#333333")) +
  labs(
    title = "Mean country-level obesity prevalence over time",
    subtitle = "Ages 5–19; unweighted mean across available countries",
    x = "Year",
    y = "Mean modelled prevalence (%)",
    colour = "Sex"
  )

trend_plot

ggsave("visuals/regional_obesity_trend.png", trend_plot, width = 11, height = 7, dpi = 200)

7.2 7.2 Country-level change

Absolute change is the latest minus the first estimate. Annualized change divides that difference by the number of elapsed years, allowing countries with different observation spans to be compared. Relative percentage change is reported but should be interpreted cautiously when baseline prevalence is low.

country_change <- obesity |>
  filter(Age_group == "Y5T19", Sex == "Total") |>
  group_by(Country, Country_code) |>
  arrange(Year, .by_group = TRUE) |>
  summarise(
    Start_year = first(Year),
    End_year = last(Year),
    Start_prevalence = first(Obesity_prevalence),
    End_prevalence = last(Obesity_prevalence),
    Absolute_change = End_prevalence - Start_prevalence,
    Relative_change_percent = 100 * Absolute_change / Start_prevalence,
    Annualized_change = Absolute_change / (End_year - Start_year),
    .groups = "drop"
  ) |>
  arrange(desc(Absolute_change))

kable(
  head(country_change, 15), digits = 2,
  caption = "Countries with the largest absolute increase over the available period"
)

Countries with the largest absolute increase over the available period
Country	Country_code	Start_year	End_year	Start_prevalence	End_prevalence	Absolute_change	Relative_change_percent	Annualized_change
Hungary	HUN	1990	2022	3.56	14.78	11.23	315.83	0.35
Finland	FIN	1990	2022	3.07	12.65	9.58	312.38	0.30
Türkiye	TUR	1990	2022	2.61	11.73	9.12	349.46	0.29
Estonia	EST	1990	2022	2.38	10.69	8.32	350.10	0.26
Bosnia and Herzegovina	BIH	1990	2022	3.73	12.05	8.31	222.73	0.26
Bulgaria	BGR	1990	2022	3.99	12.06	8.07	202.50	0.25
Romania	ROU	1990	2022	1.09	8.84	7.75	712.39	0.24
Cyprus	CYP	1990	2022	6.77	14.07	7.30	107.70	0.23
Slovakia	SVK	1990	2022	2.49	9.70	7.21	289.05	0.23
North Macedonia	MKD	1990	2022	4.20	11.32	7.11	169.21	0.22
Austria	AUT	1990	2022	3.58	10.57	6.98	194.82	0.22
United Kingdom of Great Britain and Northern Ireland	GBR	1990	2022	4.48	11.31	6.83	152.50	0.21
Slovenia	SVN	1990	2022	4.09	10.84	6.75	165.02	0.21
Georgia	GEO	1990	2022	3.94	10.59	6.66	169.06	0.21
Serbia	SRB	1990	2022	4.45	10.94	6.48	145.55	0.20

country_change_plot <- country_change |>
  mutate(Country = reorder(Country, Absolute_change)) |>
  ggplot(aes(Absolute_change, Country, fill = Absolute_change > 0)) +
  geom_col(show.legend = FALSE) +
  geom_vline(xintercept = 0, colour = "#333333") +
  scale_fill_manual(values = c("TRUE" = "#C94C4C", "FALSE" = "#3B78A8")) +
  labs(
    title = "Change in modelled obesity prevalence by country",
    subtitle = "First to latest available estimate; total population aged 5–19",
    x = "Absolute change (percentage points)", y = NULL
  )

country_change_plot

ggsave("visuals/country_change.png", country_change_plot, width = 11, height = 10, dpi = 200)

7.3 7.3 Did countries converge or diverge?

The P90–P10 gap is the difference between the 90th and 10th percentiles of country prevalence. The interquartile range (IQR) captures the middle half of countries and is less influenced by extremes. Both are absolute inequality measures.

country_inequality <- obesity |>
  filter(Age_group == "Y5T19", Sex == "Total") |>
  group_by(Year) |>
  summarise(
    P10 = quantile(Obesity_prevalence, 0.10, na.rm = TRUE),
    P90 = quantile(Obesity_prevalence, 0.90, na.rm = TRUE),
    P90_P10_gap = P90 - P10,
    IQR = IQR(Obesity_prevalence, na.rm = TRUE),
    Countries = n_distinct(Country),
    .groups = "drop"
  )

inequality_plot <- country_inequality |>
  select(Year, `P90–P10 gap` = P90_P10_gap, IQR) |>
  pivot_longer(-Year, names_to = "Measure", values_to = "Value") |>
  ggplot(aes(Year, Value, colour = Measure)) +
  geom_line(linewidth = 1) +
  scale_colour_manual(values = c("P90–P10 gap" = "#7A3E9D", "IQR" = "#D9795B")) +
  labs(
    title = "Between-country inequality in obesity prevalence",
    subtitle = "Total population aged 5–19; unweighted country distribution",
    x = "Year", y = "Absolute spread (percentage points)", colour = NULL
  )

inequality_plot

ggsave("visuals/between_country_inequality.png", inequality_plot, width = 11, height = 7, dpi = 200)

8 8. Uncertainty and ranking caution

Confidence-interval width is used as a transparent indicator of estimate precision. It is not a data-quality score. Countries with overlapping intervals should not be described as definitively different based on rank alone.

latest_uncertainty <- latest_total |>
  mutate(CI_width = Upper_CI - Lower_CI) |>
  arrange(desc(CI_width))

kable(
  latest_uncertainty |>
    select(Country, Obesity_prevalence, Lower_CI, Upper_CI, CI_width) |>
    head(15),
  digits = 1,
  caption = "Latest estimates with the widest confidence intervals"
)

Latest estimates with the widest confidence intervals
Country	Obesity_prevalence	Lower_CI	Upper_CI	CI_width
Andorra	11.2	3.9	21.3	17.4
Iceland	8.7	2.8	17.8	15.1
Luxembourg	7.4	2.2	15.7	13.5
Azerbaijan	8.6	3.5	16.1	12.6
Bosnia and Herzegovina	12.0	6.7	18.6	11.9
Belarus	7.9	3.3	14.2	10.8
Uzbekistan	6.4	2.3	12.8	10.4
Türkiye	11.7	7.9	16.0	8.2
Cyprus	14.1	10.5	18.0	7.5
Greenland	4.7	1.8	9.2	7.5
Georgia	10.6	7.3	14.7	7.4
Malta	12.1	8.8	15.8	6.9
North Macedonia	11.3	8.2	15.1	6.9
Armenia	9.6	6.7	13.0	6.4
Germany	8.5	5.7	11.9	6.3

9 9. Geographic patterns

Maps help reveal geographic clustering that is difficult to see in a ranked table. The map frame covers Europe and central Asia. Greenland remains in the analytical dataset but falls outside this focused display.

Country names in the WHO file are matched to the names used by the maps package through one explicit lookup table. This avoids fragile automatic text matching and creates a reusable Map_region field for Tableau.

9.1 9.1 Latest obesity prevalence across the Region

prevalence_map <- ggplot(
  prevalence_map_data,
  aes(long, lat, group = group, fill = Obesity_prevalence)
) +
  geom_polygon(colour = "white", linewidth = 0.18) +
  coord_quickmap(xlim = c(-30, 90), ylim = c(25, 75), expand = FALSE) +
  scale_fill_viridis_c(
    option = "C",
    direction = -1,
    na.value = "#E6E6E6",
    name = "Prevalence (%)"
  ) +
  labs(
    title = paste("Childhood and adolescent obesity prevalence across Europe in", latest_year),
    subtitle = "WHO modelled estimates for ages 5–19, total population",
    x = NULL,
    y = NULL,
    caption = "Darker colours indicate higher prevalence; grey indicates no matched estimate. Source: WHO"
  ) +
  theme_void(base_size = 12) +
  theme(
    legend.position = "right",
    plot.title = element_text(face = "bold"),
    plot.caption = element_text(hjust = 0)
  )

prevalence_map

ggsave(
  "visuals/latest_obesity_prevalence_map.png",
  prevalence_map,
  width = 12,
  height = 7,
  dpi = 220,
  bg = "white"
)

How to read the map: darker countries have higher modelled prevalence in 2022; grey countries are outside the matched analytical data.

Easy interpretation: Hungary has the highest point estimate at 14.8%, followed by Cyprus and Finland. The map helps identify geographic patterning, but colour differences do not by themselves prove statistically significant or causal differences.

9.2 9.2 Geographic pattern of the sex gap

max_absolute_sex_gap <- max(abs(sex_gap$Sex_gap), na.rm = TRUE)

sex_gap_map <- ggplot(
  sex_gap_map_data,
  aes(long, lat, group = group, fill = Sex_gap)
) +
  geom_polygon(colour = "white", linewidth = 0.18) +
  coord_quickmap(xlim = c(-30, 90), ylim = c(25, 75), expand = FALSE) +
  scale_fill_gradient2(
    low = "#3B78A8",
    mid = "#F7F7F7",
    high = "#C94C4C",
    midpoint = 0,
    limits = c(-max_absolute_sex_gap, max_absolute_sex_gap),
    na.value = "#E6E6E6",
    name = "Male − female\n(percentage points)"
  ) +
  labs(
    title = paste("Sex differences in obesity prevalence across Europe in", latest_year),
    subtitle = "Ages 5–19; red indicates higher prevalence among males and blue among females",
    x = NULL,
    y = NULL,
    caption = "Near-white values indicate a small sex gap; grey indicates no matched estimate. Source: WHO"
  ) +
  theme_void(base_size = 12) +
  theme(
    legend.position = "right",
    plot.title = element_text(face = "bold"),
    plot.caption = element_text(hjust = 0)
  )

sex_gap_map

ggsave(
  "visuals/latest_sex_gap_map.png",
  sex_gap_map,
  width = 12,
  height = 7,
  dpi = 220,
  bg = "white"
)

How to read the map: red means the male estimate is higher, blue means the female estimate is higher, and colours close to white indicate a small difference.

Easy interpretation: Serbia has the largest absolute sex gap (6.9 percentage points). This is a descriptive difference between modelled estimates; it does not explain whether biology, behaviour, social conditions or data inputs produced the gap.

10 10. Tableau-ready export

The export contains cleaned fields plus latest-year flags, quartiles and sex-gap values. These derived variables simplify Tableau dashboard construction.

tableau_export <- obesity |>
  left_join(map_name_lookup, by = "Country") |>
  left_join(
    sex_gap |>
      select(Country_code, Sex_gap),
    by = "Country_code"
  ) |>
  left_join(
    latest_total |>
      select(Country_code, Burden_quartile),
    by = "Country_code"
  ) |>
  left_join(
    country_change |>
      select(Country_code, Absolute_change, Relative_change_percent, Annualized_change),
    by = "Country_code"
  ) |>
  mutate(
    CI_width = Upper_CI - Lower_CI,
    Latest_year = latest_year,
    Is_latest_year = Year == latest_year,
    Latest_burden_quartile = ifelse(
      Year == latest_year & Sex == "Total" & Age_group == "Y5T19",
      as.character(Burden_quartile),
      NA_character_
    )
  ) |>
  select(-Burden_quartile)

write.csv(
  tableau_export,
  "data/processed/tableau_childhood_obesity.csv",
  row.names = FALSE,
  na = ""
)

kable(
  head(tableau_export, 8),
  digits = 2,
  caption = "Preview of the Tableau-ready dataset"
)

Preview of the Tableau-ready dataset
Country	Country_code	Year	Sex	Age_group	Age_group_label	Obesity_prevalence	Lower_CI	Upper_CI	Source	Map_region	Sex_gap	Absolute_change	Relative_change_percent	Annualized_change	CI_width	Latest_year	Is_latest_year	Latest_burden_quartile
Albania	ALB	1990	Female	Y10T19	10–19 years	1.28	0.35	3.06	https://data.who.int/indicators/i/C6262EC/EF93DDB	Albania	4.27	5	198.31	0.16	2.71	2022	FALSE	NA
Albania	ALB	1990	Female	Y5T19	5–19 years	1.69	0.49	3.89	https://data.who.int/indicators/i/C6262EC/EF93DDB	Albania	4.27	5	198.31	0.16	3.39	2022	FALSE	NA
Albania	ALB	1990	Female	Y5T9	5–9 years	2.45	0.74	5.54	https://data.who.int/indicators/i/C6262EC/EF93DDB	Albania	4.27	5	198.31	0.16	4.80	2022	FALSE	NA
Albania	ALB	1990	Male	Y10T19	10–19 years	2.24	0.66	5.24	https://data.who.int/indicators/i/C6262EC/EF93DDB	Albania	4.27	5	198.31	0.16	4.58	2022	FALSE	NA
Albania	ALB	1990	Male	Y5T19	5–19 years	3.32	1.09	7.49	https://data.who.int/indicators/i/C6262EC/EF93DDB	Albania	4.27	5	198.31	0.16	6.40	2022	FALSE	NA
Albania	ALB	1990	Male	Y5T9	5–9 years	5.29	1.78	11.46	https://data.who.int/indicators/i/C6262EC/EF93DDB	Albania	4.27	5	198.31	0.16	9.68	2022	FALSE	NA
Albania	ALB	1990	Total	Y10T19	10–19 years	1.77	0.75	3.49	https://data.who.int/indicators/i/C6262EC/EF93DDB	Albania	4.27	5	198.31	0.16	2.74	2022	FALSE	NA
Albania	ALB	1990	Total	Y5T19	5–19 years	2.52	1.10	4.79	https://data.who.int/indicators/i/C6262EC/EF93DDB	Albania	4.27	5	198.31	0.16	3.69	2022	FALSE	NA

11 11. Interpretation and public-health relevance

This is a surveillance and inequality-screening analysis. A high or rapidly increasing estimate identifies where closer investigation may be warranted; it does not identify why prevalence is high. Confidence intervals and changes in the country distribution are more informative than a league table alone.

The conference report and WHO evidence suggest a practical interpretation pathway:

Surveillance: compare modelled patterns with standardized measured systems such as COSI and national child-growth registers.
Equity diagnosis: add HBSC family-affluence, diet and physical-activity indicators before making claims about socioeconomic inequity.
Systems response: examine school meals, active travel, physical education, food retail and digital marketing together rather than treating behaviour as an isolated choice.
Policy audit: link country trends to comparable policy-implementation data before evaluating marketing restrictions or other interventions.

These steps convert descriptive findings into hypotheses for further study; they are not causal conclusions from the current data.

12 12. Limitations

The values are modelled estimates rather than a single harmonised survey.
Country comparisons may reflect differences in source data and uncertainty.
The regional trend is not population weighted.
Country-level associations cannot be applied to individuals.
Prevalence alone does not measure access to prevention or treatment.
Statistical overlap between confidence intervals should be considered before describing differences as meaningful.
The data end in 2022 and therefore do not measure subsequent changes reported by COSI 2022–2024.
Sex is represented as Female, Male and Total; gender identity and intersectional inequalities cannot be assessed.
P90–P10 and IQR summarize between-country dispersion but do not measure within-country social inequality.
No correction is made for multiple comparisons, and ranks are descriptive.

13 13. Recommended next-stage data linkage

To investigate mechanisms while maintaining a defensible design, a second phase should link the WHO prevalence series to country-year indicators from:

HBSC: family affluence, diet, physical activity and self-reported weight status;
COSI: standardized measured prevalence, family behaviours and school environments for children aged 6–9;
WHO policy sources: implementation of food-marketing restrictions, school food policies and nutrition actions; and
Eurostat or World Bank: material deprivation, income inequality and urbanization.

The primary extension question would be: Are country-level obesity trends associated with socioeconomic conditions and policy environments after accounting for time and repeated observations? A panel-data design with pre-specified confounders would be required. Associations would still be ecological and should not be interpreted as individual causal effects.

14 14. Reproducibility checklist

Keep the repository folder structure unchanged.
Do not edit the raw CSV manually.
Install the packages listed in required_packages.
Open childhood-obesity-analysis.Rmd in RStudio.
Set the repository root as the working project folder.
Click Knit to create the HTML report.
Confirm that visuals/ and data/processed/ are regenerated.

15 15. Session information

sessionInfo()

## R version 4.5.2 (2025-10-31)
## Platform: aarch64-apple-darwin20
## Running under: macOS Tahoe 26.5.2
## 
## Matrix products: default
## BLAS:   /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib 
## LAPACK: /Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/lib/libRlapack.dylib;  LAPACK version 3.12.1
## 
## locale:
## [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
## 
## time zone: Europe/Helsinki
## tzcode source: internal
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
## [1] tidyr_1.3.2   maps_3.4.3    knitr_1.51    ggplot2_4.0.3 dplyr_1.2.1  
## 
## loaded via a namespace (and not attached):
##  [1] gtable_0.3.6       jsonlite_2.0.0     compiler_4.5.2     tidyselect_1.2.1  
##  [5] jquerylib_0.1.4    scales_1.4.0       yaml_2.3.12        fastmap_1.2.0     
##  [9] R6_2.6.1           labeling_0.4.3     generics_0.1.4     tibble_3.3.1      
## [13] bslib_0.12.0       pillar_1.11.1      RColorBrewer_1.1-3 rlang_1.3.0       
## [17] cachem_1.1.0       xfun_0.60          sass_0.4.10        S7_0.2.2          
## [21] otel_0.2.0         viridisLite_0.4.3  cli_3.6.6          withr_3.0.3       
## [25] magrittr_2.0.5     digest_0.6.39      grid_4.5.2         lifecycle_1.0.5   
## [29] vctrs_0.7.3        evaluate_1.0.5     glue_1.8.1         farver_2.1.2      
## [33] rmarkdown_2.31     purrr_1.2.2        tools_4.5.2        pkgconfig_2.0.3   
## [37] htmltools_0.5.9

European Childhood Obesity Inequalities

Burden, trends, sex gaps and surveillance uncertainty across the WHO European Region

Md. Azmir Alam

11 August 2026